fs/btrfs/root-tree.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/err.h>
   7 #include <linux/uuid.h>
   8 #include "ctree.h"
   9 #include "transaction.h"
  10 #include "disk-io.h"
  11 #include "print-tree.h"
  12 #include "qgroup.h"
  13 #include "space-info.h"
  14
  15 /*
  16  * Read a root item from the tree. In case we detect a root item smaller then
  17  * sizeof(root_item), we know it's an old version of the root structure and
  18  * initialize all new fields to zero. The same happens if we detect mismatching
  19  * generation numbers as then we know the root was once mounted with an older
  20  * kernel that was not aware of the root item structure change.
  21  */
  22 static void btrfs_read_root_item(struct extent_buffer *eb, int slot,
  23                                 struct btrfs_root_item *item)
  24 {
  25         u32 len;
  26         int need_reset = 0;
  27
  28         len = btrfs_item_size_nr(eb, slot);
  29         read_extent_buffer(eb, item, btrfs_item_ptr_offset(eb, slot),
  30                            min_t(u32, len, sizeof(*item)));
  31         if (len < sizeof(*item))
  32                 need_reset = 1;
  33         if (!need_reset && btrfs_root_generation(item)
  34                 != btrfs_root_generation_v2(item)) {
  35                 if (btrfs_root_generation_v2(item) != 0) {
  36                         btrfs_warn(eb->fs_info,
  37                                         "mismatching generation and generation_v2 found in root item. This root was probably mounted with an older kernel. Resetting all new fields.");
  38                 }
  39                 need_reset = 1;
  40         }
  41         if (need_reset) {
  42                 memset(&item->generation_v2, 0,
  43                         sizeof(*item) - offsetof(struct btrfs_root_item,
  44                                         generation_v2));
  45
  46                 generate_random_guid(item->uuid);
  47         }
  48 }
  49
  50 /*
  51  * btrfs_find_root - lookup the root by the key.
  52  * root: the root of the root tree
  53  * search_key: the key to search
  54  * path: the path we search
  55  * root_item: the root item of the tree we look for
  56  * root_key: the root key of the tree we look for
  57  *
  58  * If ->offset of 'search_key' is -1ULL, it means we are not sure the offset
  59  * of the search key, just lookup the root with the highest offset for a
  60  * given objectid.
  61  *
  62  * If we find something return 0, otherwise > 0, < 0 on error.
  63  */
  64 int btrfs_find_root(struct btrfs_root *root, const struct btrfs_key *search_key,
  65                     struct btrfs_path *path, struct btrfs_root_item *root_item,
  66                     struct btrfs_key *root_key)
  67 {
  68         struct btrfs_key found_key;
  69         struct extent_buffer *l;
  70         int ret;
  71         int slot;
  72
  73         ret = btrfs_search_slot(NULL, root, search_key, path, 0, 0);
  74         if (ret < 0)
  75                 return ret;
  76
  77         if (search_key->offset != -1ULL) {      /* the search key is exact */
  78                 if (ret > 0)
  79                         goto out;
  80         } else {
  81                 BUG_ON(ret == 0);               /* Logical error */
  82                 if (path->slots[0] == 0)
  83                         goto out;
  84                 path->slots[0]--;
  85                 ret = 0;
  86         }
  87
  88         l = path->nodes[0];
  89         slot = path->slots[0];
  90
  91         btrfs_item_key_to_cpu(l, &found_key, slot);
  92         if (found_key.objectid != search_key->objectid ||
  93             found_key.type != BTRFS_ROOT_ITEM_KEY) {
  94                 ret = 1;
  95                 goto out;
  96         }
  97
  98         if (root_item)
  99                 btrfs_read_root_item(l, slot, root_item);
 100         if (root_key)
 101                 memcpy(root_key, &found_key, sizeof(found_key));
 102 out:
 103         btrfs_release_path(path);
 104         return ret;
 105 }
 106
 107 void btrfs_set_root_node(struct btrfs_root_item *item,
 108                          struct extent_buffer *node)
 109 {
 110         btrfs_set_root_bytenr(item, node->start);
 111         btrfs_set_root_level(item, btrfs_header_level(node));
 112         btrfs_set_root_generation(item, btrfs_header_generation(node));
 113 }
 114
 115 /*
 116  * copy the data in 'item' into the btree
 117  */
 118 int btrfs_update_root(struct btrfs_trans_handle *trans, struct btrfs_root
 119                       *root, struct btrfs_key *key, struct btrfs_root_item
 120                       *item)
 121 {
 122         struct btrfs_fs_info *fs_info = root->fs_info;
 123         struct btrfs_path *path;
 124         struct extent_buffer *l;
 125         int ret;
 126         int slot;
 127         unsigned long ptr;
 128         u32 old_len;
 129
 130         path = btrfs_alloc_path();
 131         if (!path)
 132                 return -ENOMEM;
 133
 134         ret = btrfs_search_slot(trans, root, key, path, 0, 1);
 135         if (ret < 0)
 136                 goto out;
 137
 138         if (ret > 0) {
 139                 btrfs_crit(fs_info,
 140                         "unable to find root key (%llu %u %llu) in tree %llu",
 141                         key->objectid, key->type, key->offset,
 142                         root->root_key.objectid);
 143                 ret = -EUCLEAN;
 144                 btrfs_abort_transaction(trans, ret);
 145                 goto out;
 146         }
 147
 148         l = path->nodes[0];
 149         slot = path->slots[0];
 150         ptr = btrfs_item_ptr_offset(l, slot);
 151         old_len = btrfs_item_size_nr(l, slot);
 152
 153         /*
 154          * If this is the first time we update the root item which originated
 155          * from an older kernel, we need to enlarge the item size to make room
 156          * for the added fields.
 157          */
 158         if (old_len < sizeof(*item)) {
 159                 btrfs_release_path(path);
 160                 ret = btrfs_search_slot(trans, root, key, path,
 161                                 -1, 1);
 162                 if (ret < 0) {
 163                         btrfs_abort_transaction(trans, ret);
 164                         goto out;
 165                 }
 166
 167                 ret = btrfs_del_item(trans, root, path);
 168                 if (ret < 0) {
 169                         btrfs_abort_transaction(trans, ret);
 170                         goto out;
 171                 }
 172                 btrfs_release_path(path);
 173                 ret = btrfs_insert_empty_item(trans, root, path,
 174                                 key, sizeof(*item));
 175                 if (ret < 0) {
 176                         btrfs_abort_transaction(trans, ret);
 177                         goto out;
 178                 }
 179                 l = path->nodes[0];
 180                 slot = path->slots[0];
 181                 ptr = btrfs_item_ptr_offset(l, slot);
 182         }
 183
 184         /*
 185          * Update generation_v2 so at the next mount we know the new root
 186          * fields are valid.
 187          */
 188         btrfs_set_root_generation_v2(item, btrfs_root_generation(item));
 189
 190         write_extent_buffer(l, item, ptr, sizeof(*item));
 191         btrfs_mark_buffer_dirty(path->nodes[0]);
 192 out:
 193         btrfs_free_path(path);
 194         return ret;
 195 }
 196
 197 int btrfs_insert_root(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 198                       const struct btrfs_key *key, struct btrfs_root_item *item)
 199 {
 200         /*
 201          * Make sure generation v1 and v2 match. See update_root for details.
 202          */
 203         btrfs_set_root_generation_v2(item, btrfs_root_generation(item));
 204         return btrfs_insert_item(trans, root, key, item, sizeof(*item));
 205 }
 206
 207 int btrfs_find_orphan_roots(struct btrfs_fs_info *fs_info)
 208 {
 209         struct btrfs_root *tree_root = fs_info->tree_root;
 210         struct extent_buffer *leaf;
 211         struct btrfs_path *path;
 212         struct btrfs_key key;
 213         struct btrfs_key root_key;
 214         struct btrfs_root *root;
 215         int err = 0;
 216         int ret;
 217
 218         path = btrfs_alloc_path();
 219         if (!path)
 220                 return -ENOMEM;
 221
 222         key.objectid = BTRFS_ORPHAN_OBJECTID;
 223         key.type = BTRFS_ORPHAN_ITEM_KEY;
 224         key.offset = 0;
 225
 226         root_key.type = BTRFS_ROOT_ITEM_KEY;
 227         root_key.offset = (u64)-1;
 228
 229         while (1) {
 230                 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
 231                 if (ret < 0) {
 232                         err = ret;
 233                         break;
 234                 }
 235
 236                 leaf = path->nodes[0];
 237                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 238                         ret = btrfs_next_leaf(tree_root, path);
 239                         if (ret < 0)
 240                                 err = ret;
 241                         if (ret != 0)
 242                                 break;
 243                         leaf = path->nodes[0];
 244                 }
 245
 246                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 247                 btrfs_release_path(path);
 248
 249                 if (key.objectid != BTRFS_ORPHAN_OBJECTID ||
 250                     key.type != BTRFS_ORPHAN_ITEM_KEY)
 251                         break;
 252
 253                 root_key.objectid = key.offset;
 254                 key.offset++;
 255
 256                 root = btrfs_get_fs_root(fs_info, &root_key, false);
 257                 err = PTR_ERR_OR_ZERO(root);
 258                 if (err && err != -ENOENT) {
 259                         break;
 260                 } else if (err == -ENOENT) {
 261                         struct btrfs_trans_handle *trans;
 262
 263                         btrfs_release_path(path);
 264
 265                         trans = btrfs_join_transaction(tree_root);
 266                         if (IS_ERR(trans)) {
 267                                 err = PTR_ERR(trans);
 268                                 btrfs_handle_fs_error(fs_info, err,
 269                                             "Failed to start trans to delete orphan item");
 270                                 break;
 271                         }
 272                         err = btrfs_del_orphan_item(trans, tree_root,
 273                                                     root_key.objectid);
 274                         btrfs_end_transaction(trans);
 275                         if (err) {
 276                                 btrfs_handle_fs_error(fs_info, err,
 277                                             "Failed to delete root orphan item");
 278                                 break;
 279                         }
 280                         continue;
 281                 }
 282
 283                 WARN_ON(!test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state));
 284                 if (btrfs_root_refs(&root->root_item) == 0) {
 285                         set_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
 286                         btrfs_add_dead_root(root);
 287                 }
 288                 btrfs_put_root(root);
 289         }
 290
 291         btrfs_free_path(path);
 292         return err;
 293 }
 294
 295 /* drop the root item for 'key' from the tree root */
 296 int btrfs_del_root(struct btrfs_trans_handle *trans,
 297                    const struct btrfs_key *key)
 298 {
 299         struct btrfs_root *root = trans->fs_info->tree_root;
 300         struct btrfs_path *path;
 301         int ret;
 302
 303         path = btrfs_alloc_path();
 304         if (!path)
 305                 return -ENOMEM;
 306         ret = btrfs_search_slot(trans, root, key, path, -1, 1);
 307         if (ret < 0)
 308                 goto out;
 309
 310         BUG_ON(ret != 0);
 311
 312         ret = btrfs_del_item(trans, root, path);
 313 out:
 314         btrfs_free_path(path);
 315         return ret;
 316 }
 317
 318 int btrfs_del_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
 319                        u64 ref_id, u64 dirid, u64 *sequence, const char *name,
 320                        int name_len)
 321
 322 {
 323         struct btrfs_root *tree_root = trans->fs_info->tree_root;
 324         struct btrfs_path *path;
 325         struct btrfs_root_ref *ref;
 326         struct extent_buffer *leaf;
 327         struct btrfs_key key;
 328         unsigned long ptr;
 329         int err = 0;
 330         int ret;
 331
 332         path = btrfs_alloc_path();
 333         if (!path)
 334                 return -ENOMEM;
 335
 336         key.objectid = root_id;
 337         key.type = BTRFS_ROOT_BACKREF_KEY;
 338         key.offset = ref_id;
 339 again:
 340         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
 341         BUG_ON(ret < 0);
 342         if (ret == 0) {
 343                 leaf = path->nodes[0];
 344                 ref = btrfs_item_ptr(leaf, path->slots[0],
 345                                      struct btrfs_root_ref);
 346                 ptr = (unsigned long)(ref + 1);
 347                 if ((btrfs_root_ref_dirid(leaf, ref) != dirid) ||
 348                     (btrfs_root_ref_name_len(leaf, ref) != name_len) ||
 349                     memcmp_extent_buffer(leaf, name, ptr, name_len)) {
 350                         err = -ENOENT;
 351                         goto out;
 352                 }
 353                 *sequence = btrfs_root_ref_sequence(leaf, ref);
 354
 355                 ret = btrfs_del_item(trans, tree_root, path);
 356                 if (ret) {
 357                         err = ret;
 358                         goto out;
 359                 }
 360         } else
 361                 err = -ENOENT;
 362
 363         if (key.type == BTRFS_ROOT_BACKREF_KEY) {
 364                 btrfs_release_path(path);
 365                 key.objectid = ref_id;
 366                 key.type = BTRFS_ROOT_REF_KEY;
 367                 key.offset = root_id;
 368                 goto again;
 369         }
 370
 371 out:
 372         btrfs_free_path(path);
 373         return err;
 374 }
 375
 376 /*
 377  * add a btrfs_root_ref item.  type is either BTRFS_ROOT_REF_KEY
 378  * or BTRFS_ROOT_BACKREF_KEY.
 379  *
 380  * The dirid, sequence, name and name_len refer to the directory entry
 381  * that is referencing the root.
 382  *
 383  * For a forward ref, the root_id is the id of the tree referencing
 384  * the root and ref_id is the id of the subvol  or snapshot.
 385  *
 386  * For a back ref the root_id is the id of the subvol or snapshot and
 387  * ref_id is the id of the tree referencing it.
 388  *
 389  * Will return 0, -ENOMEM, or anything from the CoW path
 390  */
 391 int btrfs_add_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
 392                        u64 ref_id, u64 dirid, u64 sequence, const char *name,
 393                        int name_len)
 394 {
 395         struct btrfs_root *tree_root = trans->fs_info->tree_root;
 396         struct btrfs_key key;
 397         int ret;
 398         struct btrfs_path *path;
 399         struct btrfs_root_ref *ref;
 400         struct extent_buffer *leaf;
 401         unsigned long ptr;
 402
 403         path = btrfs_alloc_path();
 404         if (!path)
 405                 return -ENOMEM;
 406
 407         key.objectid = root_id;
 408         key.type = BTRFS_ROOT_BACKREF_KEY;
 409         key.offset = ref_id;
 410 again:
 411         ret = btrfs_insert_empty_item(trans, tree_root, path, &key,
 412                                       sizeof(*ref) + name_len);
 413         if (ret) {
 414                 btrfs_abort_transaction(trans, ret);
 415                 btrfs_free_path(path);
 416                 return ret;
 417         }
 418
 419         leaf = path->nodes[0];
 420         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
 421         btrfs_set_root_ref_dirid(leaf, ref, dirid);
 422         btrfs_set_root_ref_sequence(leaf, ref, sequence);
 423         btrfs_set_root_ref_name_len(leaf, ref, name_len);
 424         ptr = (unsigned long)(ref + 1);
 425         write_extent_buffer(leaf, name, ptr, name_len);
 426         btrfs_mark_buffer_dirty(leaf);
 427
 428         if (key.type == BTRFS_ROOT_BACKREF_KEY) {
 429                 btrfs_release_path(path);
 430                 key.objectid = ref_id;
 431                 key.type = BTRFS_ROOT_REF_KEY;
 432                 key.offset = root_id;
 433                 goto again;
 434         }
 435
 436         btrfs_free_path(path);
 437         return 0;
 438 }
 439
 440 /*
 441  * Old btrfs forgets to init root_item->flags and root_item->byte_limit
 442  * for subvolumes. To work around this problem, we steal a bit from
 443  * root_item->inode_item->flags, and use it to indicate if those fields
 444  * have been properly initialized.
 445  */
 446 void btrfs_check_and_init_root_item(struct btrfs_root_item *root_item)
 447 {
 448         u64 inode_flags = btrfs_stack_inode_flags(&root_item->inode);
 449
 450         if (!(inode_flags & BTRFS_INODE_ROOT_ITEM_INIT)) {
 451                 inode_flags |= BTRFS_INODE_ROOT_ITEM_INIT;
 452                 btrfs_set_stack_inode_flags(&root_item->inode, inode_flags);
 453                 btrfs_set_root_flags(root_item, 0);
 454                 btrfs_set_root_limit(root_item, 0);
 455         }
 456 }
 457
 458 void btrfs_update_root_times(struct btrfs_trans_handle *trans,
 459                              struct btrfs_root *root)
 460 {
 461         struct btrfs_root_item *item = &root->root_item;
 462         struct timespec64 ct;
 463
 464         ktime_get_real_ts64(&ct);
 465         spin_lock(&root->root_item_lock);
 466         btrfs_set_root_ctransid(item, trans->transid);
 467         btrfs_set_stack_timespec_sec(&item->ctime, ct.tv_sec);
 468         btrfs_set_stack_timespec_nsec(&item->ctime, ct.tv_nsec);
 469         spin_unlock(&root->root_item_lock);
 470 }
 471
 472 /*
 473  * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
 474  * root: the root of the parent directory
 475  * rsv: block reservation
 476  * items: the number of items that we need do reservation
 477  * use_global_rsv: allow fallback to the global block reservation
 478  *
 479  * This function is used to reserve the space for snapshot/subvolume
 480  * creation and deletion. Those operations are different with the
 481  * common file/directory operations, they change two fs/file trees
 482  * and root tree, the number of items that the qgroup reserves is
 483  * different with the free space reservation. So we can not use
 484  * the space reservation mechanism in start_transaction().
 485  */
 486 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
 487                                      struct btrfs_block_rsv *rsv, int items,
 488                                      bool use_global_rsv)
 489 {
 490         u64 qgroup_num_bytes = 0;
 491         u64 num_bytes;
 492         int ret;
 493         struct btrfs_fs_info *fs_info = root->fs_info;
 494         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
 495
 496         if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
 497                 /* One for parent inode, two for dir entries */
 498                 qgroup_num_bytes = 3 * fs_info->nodesize;
 499                 ret = btrfs_qgroup_reserve_meta_prealloc(root,
 500                                 qgroup_num_bytes, true);
 501                 if (ret)
 502                         return ret;
 503         }
 504
 505         num_bytes = btrfs_calc_insert_metadata_size(fs_info, items);
 506         rsv->space_info = btrfs_find_space_info(fs_info,
 507                                             BTRFS_BLOCK_GROUP_METADATA);
 508         ret = btrfs_block_rsv_add(root, rsv, num_bytes,
 509                                   BTRFS_RESERVE_FLUSH_ALL);
 510
 511         if (ret == -ENOSPC && use_global_rsv)
 512                 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true);
 513
 514         if (ret && qgroup_num_bytes)
 515                 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
 516
 517         return ret;
 518 }
 519
 520 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
 521                                       struct btrfs_block_rsv *rsv)
 522 {
 523         btrfs_block_rsv_release(fs_info, rsv, (u64)-1, NULL);
 524 }